As computer systems evolve, so does the demand for increased memory for such systems. To increase memory density, some memory modules stack integrated circuit (IC) dies one on top of the other. While memory subsystems commonly use die-stacking, System-in-Package (SIP) systems may also include stacked IC processor and controller die. These stacked systems permit high IC densities, thereby increasing the memory capacity of each module without requiring additional space on the underlying circuit board. Die stacking, however, does present a number of drawbacks, as described below.
In these stacked systems, the bare silicon die are typically given an overcoat of oxide to protect the die during handling. A redistribution layer (RDL) of metal may then be deposited on top of this oxide to form an external interconnection system. Holes or contacts are then etched in the oxide so the RDL metal can connect to the internal metal layers of the silicon die. When the silicon die are assembled into a vertical stack, the RDLs allow signals to pass through the stack.
Such RDLs may be appropriate for bussed (multi-drop) connections, where all of the silicon die in a stack are coupled to the same bus. However, such RDL systems are not well suited to point-to-point connections, where separate connections need to be made to individual die in the stack. This is because point-to-point connections typically require complex and custom RDLs on each die to properly route the signals through the stack. These custom RDLs on each silicon die are complex and costly to design and manufacture, particularly in the case in which all the silicon die are the same (e.g., memory die). Accordingly, a system that eliminates custom RDLs in a stacked system would be highly desirable.